Variegated competing peer-to-peer systems with selfish peers

Thanks to years of research and development, current peer-to-peer (P2P) networks are anything but a homogeneous system from a protocol perspective. Specifically, even for the same P2P system (e.g., BitTorrent), a large number of protocol variants have been designed based on game theoretic considerations with the objective to gain performance advantages. We envision that such variants could be deployed by selfish participants and interact with the original prescribed protocol as well as among them. Consequently, a meta-strategic situation-judicious selection of different protocol variants-will emerge. In this work, we propose the usage of population games, evolutionary and learning dynamics in the study of node coevolution for selfish protocol selection, and, most importantly, its impact on system performance. We apply our models and algorithms to P2P systems and draw on extensive simulations to characterize the dynamics of selfish protocol selection. In particular, our proposed distributed algorithms outperform others in terms of download rate. We believe that evolution patterns identified in our study will shed light on both further theoretical study and the design of next generation distributed systems.